A New Algorithm Of Extraction And Quantization And Coding Based On Mixed Excitation Parameters With Its Application

With the rapid development of communication technology, reduce the rate of transmission of digital become a key problem. In order to reduce transport costs, must to compress the speech signal which is obtained by d/a conversion technology. That prompted the development of speech compression coding technology. One main purpose of speech compression is expressing signal from the source as a minimum number of bits, reduce redundancy, improve the efficiency of transmission.In low rate speech coding, mixed Excitation Linear predictive coding（Mixed Excitation Linear Prediction, MELP） became one of the important research topic. MELP introduced five improved characteristics based on the LPC vocoder,five improved features are mixed pulse and noise excitation, periodic or aperiodic pulse, adaptive spectral enhancement, pulse walk filter and residual error harmonic amplitude model. Among them, the most important features of MELP is using mixed pulse and noise excitation as excitation signal.Concrete is implemented by a multi-band mixed model,the model adopts the subband unvoiced/voiced intensity modulation five band filter, gets the shaping filter,so as to gets more accurate information about the incentive.This paper proposes a new parameter of subband voiced intensity and extraction algorithm of the parameter,for many multiplications in the mixed excitation parameter extraction algorithm, calculation time consuming,five subband bandpass unvoiced/ voiced adopts quantitative problems such as a hard decisions.This algorithm adopts linear combination（α（?）+βMn） of the average period amplitude（are recorded as （?）） of short time average magnitude difference function and short time average amplitude（are recordedas M_n） token subband unvoiced/voiced intensity,the combination coefficienta and β be determined by experiment.Each dimension data of the new parameter is a number between 0¹, are recorded as P_zn.First of all,The experiment divides the subband for each frame speech signal use filter group,then from each subband signal extracts （?） and M_n,and （?） and M_n are combinedaccording to the combination coefficient α an β were determined experimentally,as a unvoiced/voiced intensity P_zn of each subband,at the same time to carry on the design to quantizer of the parameter P_zn,with the improved LBG algorithm to carry on training to codebook;and introduces the search algorithm of codebook in detail;finally use the parameter P_zn replace part of subband unvoiced/voiced intensity extraction of vocoder,the rest remains the same,done lots of simulation experiments,respectively get the vocoder based on the original subband subband unvoiced/voiced intensity and the vocoder based on the parameter P_zn.the experimental results show that decoding speech of MELP vocoder based on new parameter P_zn not only have the same speech quality with the decoding speech of original MELP vocoder, but also reduces the time complexity and improves the operation speed of the algorithm.